Architecture at scale
Setting technical direction for complex platforms and high-traffic systems, with clear boundaries and standards that let teams move independently.
I lead architecture across large-scale systems and grow engineering teams that deliver ambitious products. I am a French and British citizen based in Paris, with more than fifteen years of experience.
Technical leadership at the intersection of systems architecture, organisation-scale delivery and engineer growth — setting direction, creating leverage and keeping product outcomes in view.
Setting technical direction for complex platforms and high-traffic systems, with clear boundaries and standards that let teams move independently.
Growing engineering groups, developing staff engineers and building the structures and ownership that turn strong individuals into durable teams.
Connecting architecture and delivery to customer needs, risk and business outcomes — so technical investment earns its place.
Making coding agents useful in everyday engineering through reusable skills, evaluation, adoption and workflows that improve over time.
co-led shipping of the Wearables Device Access Toolkit, the first SDK opening Meta's AI glasses to third-party developers
customers migrated through Secure Customer Authentication in three months at Qonto
engineers across 5 teams mentored, taught and supported as an individual contributor at Meta
I work at the cutting edge of using and improving AI coding agents to help product, design and engineering teams become more productive. As a heavy daily user, I bring Meta's latest internal AI efforts to the team — improving the efficiency and effectiveness of AI across their projects through reusable skills, evaluation, and repeatable engineering workflows.
I manage the performance and evaluation of these systems over time, using evidence from real-world use to improve reliability and impact, and I lead adoption so product, design and engineering use AI coding agents effectively in their daily work.
Externally, I define how third-party developers bring AI to the platform — llms.txt, MCP servers and agent-ready documentation that let outside teams build with agents, not just APIs — alongside my own AI-native projects.
I oversee architecture across the Wearables Device Access Toolkit — the first SDK opening Meta's AI glasses to third-party developers building hands-free experiences — working across 60+ engineers in five teams as an individual contributor, alongside the director of engineering.
The platform opens access to the camera, microphone, speaker and on-device display. Much of my work sits where AI meets wearables, with a particular focus on the edge.
Grew the engineering group focused on acquisition flows at French fintech Qonto from four to eighteen engineers across three teams, pairing product delivery with the technical and organisational work needed to scale.
Led the Secure Customer Authentication programme: migrated every internal API to asynchronous behaviour, then shipped biometric confirmation on top — with no downtime for existing products and minimal interruption for users. 200,000 customers migrated in three months, fraud reduced.
Owned onboarding architecture: a banking signup flow operating under four European countries' banking laws, with KBIS and KYC verification — continuously simplified and tuned to cut dropout.
Played a key role in moving 150,000 customer accounts from a partner core-banking system onto Qonto’s own platform.
Trained staff engineers, built shared Kafka libraries, and raised the bar for asynchronous worker reliability, observability and data pipelines across the organisation.
Ran a team of five developers alongside a product manager and designer, using lean development to balance long-term company goals with short, quick iterations.
Designed a new car-data architecture that decoupled the Elm front end from the Rails back end, enabling parallel delivery and reuse across other product problems.
Introduced Clojure for data collection, building adoption through individual engagement, a week-long spike and workshops for junior and senior developers.
Developed a CQRS architecture for third-party data feeds and a front-end API using Clojure, PostgreSQL and Elasticsearch.
Built the Boolean Tree Query API that powered search, filtering and sorting across uSwitch Mobiles pages and associated services.
Built tracking pipelines into BigQuery and worked on the Rails client portal, while helping move the wider architecture into uSwitch’s internal Kubernetes cluster without downtime.
Developed a Ruby service-oriented architecture spanning Rails, Sidekiq and Sinatra, with AngularJS interfaces and Elasticsearch-backed search.
Improved production monitoring with New Relic, Logstash and Sensu, and built fuzzy-search systems using Elasticsearch and RabbitMQ.
Rewrote the tracking pipeline in Node.js while contributing across AWS infrastructure, continuous deployment, technical planning and project delivery.
Built payment-processing software in Python with PayPal and Braintree, alongside content-management systems using Pylons, Pyramid, PostgreSQL and MySQL.
Developed the BlueJ and Greenfoot Java IDEs, and built websites for students and teachers in Ruby on Rails.
Worked on visualisations of user activity and database research into mistakes made by novice programmers.
Worked in the Agile development lifecycle of Gigaset WiMAX routers, focusing on testing and bug fixing, and designed a Java tool for testing multiple routers.
University of Kent
The model is not the product. The harness is.
Coding agents — Muse Code, Claude Code, Codex — have commoditised the model. What separates teams that ship with agents from teams that drown in them is everything around it: skills that encode how your team actually works, evals that catch breakage before users do, MCP harnesses that give agents real tools with real guardrails.
I work both sides of this. Inside Meta I take the company's latest AI — Muse-family models and internal tooling — and turn it into reusable skills, measured evals and repeatable workflows that product, design and engineering teams genuinely adopt. Externally, I define how third-party developers bring AI to the Wearables platform: llms.txt, MCP servers, agent-ready documentation.
Agents don't replace engineering judgment — they amplify the teams that have it. My work is making that amplification reliable: define what good looks like, measure it, gate on it.
Muse Code · Claude Code · Codex · MCP · llms.txt · Skills · Evals · Agent harnesses · Llama Stack
MySQL, PostgreSQL, Elasticsearch
Rust, Go, Haskell, Kotlin, C++, Swift, PHP/Hack, Clojure, Ruby (Rails and Sinatra), Python (Pyramid and Pylons), Java, JavaScript (Vanilla, NodeJS and Angular)
Apache, Nginx, Chef, Puppet, Terraform, AWS, Google Cloud, Kubernetes